D://LEAD47Local LLM Hosting
AI that never
leaves the building.
Deployment of large language models on infrastructure you control, on premise or in sovereign cloud, for work where the data cannot go to a third-party service. It is for legal, health, government, and any organisation with data that cannot leave its jurisdiction or its perimeter.
THE PROBLEM
For some organisations the genuinely useful AI work is exactly the work whose data may not be sent to someone else’s servers.
The cost of leaving this alone is rarely one visible failure. It is the slow accumulation: the workaround that became the process, the thing only one person knows, the renewal nobody questioned.
Our starting point is always the same: establish what is actually true today, then decide what to change. Work scoped against an assumption tends to solve a problem you do not have.
- 01Nobody owns itIt sits with whoever touched it last, which is not the same as being managed.
- 02No current pictureWhat you have, what it costs, and who has access are all slightly out of date.
- 03Only handled when it breaksAttention arrives after the disruption rather than before it.
WHAT YOU GET
What the engagement covers
Scoped before it starts, so you know what is included and what is not.
- 01
Feasibility first
Whether local hosting is genuinely required, or whether a contractual and technical control on a hosted service would satisfy the obligation at a fraction of the cost. We would rather tell you that than sell you hardware.
- 02
Infrastructure
Specified honestly. This needs real hardware, it has a real power and cooling requirement, and we will tell you what it costs before you commit rather than after.
- 03
Deployment
Models deployed, secured, and integrated with the systems your people already use, because an AI tool nobody can reach from their normal workflow does not get used.
- 04
Operation
Monitoring, updates, and capacity management as an ongoing service, since open models move quickly and a deployment left alone for a year falls behind.
HOW WE WORK
Discover, design, deliver, embed
Four stages with a written output at each one. You always know which stage you are in and what comes next.
- 01Weeks 1 – 2
Discover
We map how the work happens now, including the workarounds people are slightly embarrassed to mention.
- 02Weeks 3 – 4
Design
Options costed against benefit, so the choice is a decision rather than a preference.
- 03Per stage
Deliver
Built in slices that reach production and get used, each with a success measure agreed before it starts.
- 04Post-delivery
Embed
Training, documentation, and a check-in once the novelty has worn off. Adoption is the only measure that counts.
WHAT CHANGES
What you should expect
- Someone other than you owns it, with that written down.
- The current state is documented and stays documented.
- Cost is planned ahead rather than discovered at renewal.
- Decisions are made against evidence rather than assumption.
FAQ
Questions we get asked
01What is local LLM hosting?
Deployment of large language models on infrastructure you control, on premise or in sovereign cloud, for work where the data cannot go to a third-party service. It is for legal, health, government, and any organisation with data that cannot leave its jurisdiction or its perimeter.
02Do we really need to host AI ourselves, or is a private cloud tenancy enough?
For most organisations the enterprise tiers of the major services, with contractual commitments on data handling and no training on your content, satisfy the actual obligation. Genuine local hosting is warranted where a law, a government requirement, or a specific client contract says the data may not leave your control at all. That is a real category and it is smaller than the number of organisations who believe they are in it, which is the first thing we establish.
03How does a local model compare to the commercial ones?
Open models have improved enormously and are capable across summarisation, drafting, extraction, and question answering over your own documents. On the hardest reasoning tasks the leading commercial models remain ahead. For the document-grounded work that most regulated organisations actually want, the gap is usually not the deciding factor. The deciding factor is whether the data is allowed to leave.
04What hardware does this need?
Meaningful GPU capacity, with the size driven by the model and by how many people use it at once. It also needs somewhere with the power, cooling, and physical security to house it properly, which for most organisations means a data centre rather than an office cupboard. We size it against your actual expected use rather than a benchmark.
05Is this cheaper than paying per user for a commercial service?
Rarely, and anyone claiming otherwise is doing selective arithmetic. There is a capital outlay, ongoing power, and someone has to operate it. It becomes competitive at high sustained usage, and it is justified primarily by the data constraint rather than by cost. If cost is your driver, we will tell you this is the wrong answer.
06How much does local LLM hosting cost in New Zealand?
We quote after scoping rather than before. Anyone pricing this work without looking at your environment is guessing, and the guess is rarely in your favour. Scoping itself is quick, and we tell you what it costs before we start it.
07How long does it take to get started with local LLM hosting?
A first conversation takes about half an hour and costs nothing. Scoping is usually a week or two of our time depending on the size of the environment, and we agree the delivery dates with you before anything is booked in.
08Can you deliver local LLM hosting alongside our existing IT team or provider?
Yes, and it is common. We are happy to work as an extra pair of hands under your internal team, or alongside an incumbent provider on a defined piece of work. We will set out in writing where the responsibilities split, so nothing falls between us.
09Do we have to be an existing Atlas client to start a project?
No. This can be delivered as a standalone piece of work for an organisation we have never worked with before, or folded into a managed agreement if you already have one with us. Plenty of clients use us for one thing and keep everything else where it is.
10Do you deliver projects outside Auckland?
Our team is based in Auckland and we attend sites across the wider region. Most of this work is delivered remotely, so we support organisations throughout New Zealand, and we will say up front where being on site genuinely matters.
11Who from Atlas will be on the engagement?
Named people, not a queue. You get a lead who knows your environment and stays with it, which is the difference between explaining your business once and explaining it every time you make contact.
12What happens when the engagement ends?
You keep the documentation regardless, and anything registered in your name stays in your name. Whether we stay involved is your call. Some clients take it in house from there, others move it onto an ongoing agreement with us. We would rather you left cleanly than stayed because leaving was difficult.
Start with a conversation.
Tell us what you are dealing with and we will tell you whether this is the right service for it, and what it would take.
← All services